view cluster.tools/format.raw.TCGA.clinical.data.R @ 2:b442996b66ae draft

Uploaded

#!/usr/bin/env Rscript
## 
## formats raw clinical data from TCGA to contain a single status & time colums
##
## Input (required):
##    - clinical data
## Input (optional):
##    - status & time columns: (NOT USED IN THIS SCRIPT - see comment below)
##         ideally, a better design would allow a user to specify 1 or more columns
##         to check for the status & time columns - however, due to the necessities
##         required to pre-process the TCGA clinical data, the script would not be
##         generalizeable - and for this reason, the TCGA columns are hard-coded.
##
## Output: a re-formatted clinical file containing 3 columns: sample-ID, status & time
##
## Date: August 21, 2012
## Author: Peter Waltman
##

##usage, options and doc goes here
argspec <- c("format.raw.TCGA.clinical.data.R takes a clustering from ConsensusClusterPlus and clinical survival data
and generates a KM-plot, along with the log-rank p-values

        Usage: 
                format.raw.TCGA.clinical.data.R -c <clinical.file> 
        Options:
                -o <output file> (tab-delimited (3 col: sample_id <tab> status <tab> time))
              ")
args <- commandArgs(TRUE)
if ( length( args ) == 1 && args =="--help") { 
  write(argspec, stderr())
  q();
}

lib.load.quiet <- function( package ) {
   package <- as.character(substitute(package))
   suppressPackageStartupMessages( do.call( "library", list( package=package ) ) )
}
lib.load.quiet(getopt)

spec <- matrix( c( "clinical.fname", "d", 1, "character",  
                   "output.fname",   "o", 2, "character"
                  ),
                ncol=4,
                byrow=TRUE
               )
opt <- getopt( spec=spec )
save.image( "/tmp/format.dbg.rda")

##set some reasonable defaults for the options that are needed,
##but were not specified.
if ( is.null(opt$output.fname ) ) { opt$output.fname <-file.path( getwd(), "formated.TCGA.clinical.data" ) }

##orig.clinical.data <- read.delim( opt$clinical.fname, as.is=TRUE, row.names=1 )
orig.clinical.data <- read.delim( opt$clinical.fname, as.is=TRUE )
orig.clinical.data <- unique( orig.clinical.data )
rownames( orig.clinical.data ) <- orig.clinical.data[,1]
orig.clinical.data <- orig.clinical.data[, -1 ]

##  ugh, some TCGA data sets have all NAs in the "days_to_..." columns
if ( "days_to_last_known_alive" %in% colnames( orig.clinical.data ) ) {
  time.cols <- c( "days_to_death", "days_to_last_followup", "days_to_last_known_alive" )
} else {
  time.cols <- c( "days_to_death", "days_to_last_followup"  )
}
good.samps <- ! apply( orig.clinical.data[, time.cols ], 1, function(x) all( is.na(x) ) | all( x <= 0, na.rm=T ) )

orig.clinical.data <- orig.clinical.data[ good.samps, ]

if ( is.null(opt$status.column ) ) {
  status.colname <- "vital_status"
  if ( status.colname %in% colnames( orig.clinical.data ) ) {
    opt$status.column <- which( colnames( orig.clinical.data ) %in% status.colname )
    clinical.data <- orig.clinical.data[ , opt$status.column ]
  }
  else {
    status.colname <- "days_to_death"
    if ( status.colname %in% colnames( orig.clinical.data ) ) {
      opt$status.column <- which( colnames( orig.clinical.data ) %in% status.colname )
      clinical.data <- orig.clinical.data[ , opt$status.column ]
    }
    else {
      stop( "can't find a valid entry with status info - have tried vital_status & days_to_death\n" )
    }
  }
  clinical.data <- as.numeric( ! grepl( "(LIVING|Not)", clinical.data ) )
}
if ( is.null(opt$time.column ) ) {
  time.colname <- "CDE.clinical_time"
  
  if ( time.colname %in% colnames( orig.clinical.data ) ) {
    opt$time.column <- which( colnames( orig.clinical.data ) %in% time.colname )
    clinical.data <- cbind( clinical.data,
                           as.numeric( orig.clinical.data[, opt$time.column ] ) )
  }
  else {
    dec.mat <- matrix( NA,
                       nc=length( time.cols ),
                       nr=nrow( orig.clinical.data ),
                       dimnames=list( rownames( orig.clinical.data ),
                                       time.cols )
                      )
    for ( cname in colnames( dec.mat ) ) {
      if ( cname %in% colnames( orig.clinical.data ) ) {
        dec.mat[, cname ] <- as.numeric( orig.clinical.data[, cname ] )
      }
    }
                         
    

    if ( "days_to_last_known_alive" %in% colnames( orig.clinical.data ) ) {

      opt$time.column <- sapply( 1:length( clinical.data ),
                                 function(i) {
                                   if ( clinical.data[i] ) {
                                     ## this is a deceased sample
                                     return( ifelse( ( !is.na( dec.mat[ i, "days_to_death" ] ) ),
                                                     dec.mat[ i, "days_to_death" ],
                                                     ifelse( ( !is.na( dec.mat[ i, "days_to_last_known_alive" ] ) ),
                                                             dec.mat[ i, "days_to_last_known_alive" ],
                                                             dec.mat[ i, "days_to_last_followup" ] ) ) )
                                                   
                                   }
                                   else {
                                     return( max( dec.mat[ i, c( "days_to_last_followup","days_to_last_known_alive") ], na.rm=T ) )
                                   }
                                 }
                                )
    } else {
      opt$time.column <- sapply( 1:length( clinical.data ),
                                 function(i) {
                                   if ( clinical.data[i] ) {
                                     ## this is a deceased sample
                                     return( ifelse( ( !is.na( dec.mat[ i, "days_to_death" ] ) ),
                                                     dec.mat[ i, "days_to_death" ],
                                                     dec.mat[ i, "days_to_last_followup" ] ) )
                                                   
                                   }
                                   else {
                                     return( max( dec.mat[ i, c( "days_to_last_followup") ], na.rm=T ) )
                                   }
                                 }
                                )
    }
                                   
    
    clinical.data <- cbind( clinical.data,
                           as.numeric( opt$time.column ) )
  }
}

clinical.data <- as.data.frame( clinical.data )
colnames( clinical.data ) <- c( "status", "time" )
rownames( clinical.data ) <- rownames( orig.clinical.data )


##  check to make sure that the id's are sync'd correctly
## the default format is to use hyphens to separate the elt's of the name
## and to only use the 1st 3 elements of the name
## so we check to see if they're using something else as separators and/or using more than 3 elts
reformat.ids <- function( ids ) {

  if ( grepl( "TCGA\\.", ids[1] ) ) {
    ids <- sapply( strsplit( ids, "\\." ), function(x) paste( x[1:3], collapse="-" ) )
  } else {
    ## do this just in case there's more than 3 elements to the names
    if ( grepl( "TCGA-", ids[1] ) ) {
      ids <- sapply( strsplit( ids, "-" ), function(x) paste( x[1:min( c(3,length(x) ) )], collapse="-" ) )
    }
  }
  return( ids )
}


new.samp.ids <- reformat.ids( rownames( clinical.data ) )
if ( any( duplicated( new.samp.ids ) ) ) {
  ## in some cases, we have duplicate sample ids in the raw data after we truncate to
  ##   the 1st 3 elts in the barcode, so just simplify the data
  uniqs <- ! duplicated( new.samp.ids )
  clinical.data <- clinical.data[ uniqs, ]
  new.samp.ids <- new.samp.ids[ uniqs ]
}
  
rownames( clinical.data ) <- new.samp.ids
write.table( clinical.data, opt$output.fname, sep="\t", quote=FALSE, col.names=NA )